This project’s objective is to train an AI model to predict future candles of a Farex or Crypto Currency using their past data.
We trained a LSTM (long short term memory) model to learn the currency’s behavior in the targeted date range.
Read more About LSTM : WikiPedia
First we had to format the csv data to match the Tensorflow model’s desired input format for example converting the date string into datetime object and filtering the target column which is “close” in our case.
We faced some incompatibility and wide error ranges in currencies with high fundamental effects since no algorithm or model can predict such things but we managed to get more accurate results in tighter date ranges.
MSFT historical data was used as our test data.
Figure 1 is the result of our first try with a wide date span and Figure 2 is the output in a shorter time period.
In conclusion the shorter the date range the less fundamental impact a currency may face which leads to more accurate result.
In order to make this usable in real world and actually get dependable results I suggest:
1- Using this in crypto currency and shorten the date span down to minutes
2- Do not neglect the high, low and open of a candle like we did here because they have crucial impact on the next candle
